Intelligent Analysis of Prediction Uncertainty

Conveners: Mary C. Hill, US Geological Survey, and Ming Ye, Florida State University

Meaningful evaluations of prediction uncertainty require methods with the following characteristics:

They need to be computationally tractable for the long execution times characteristically needed to represent realistically natural hydrologic systems.
They need to clearly indicate what aspects of the data and model construction are dominating the calculated measures of uncertainty.
They need to indicate how to reduce uncertainty, by collecting more parameter measurements and observation data.
They need to account for different types of error, including errors in the data used for model development, errors in estimated parameters, and errors in the conceptual model.
They need to be generally applicable, and not limited to rarely satisfied conditions and assumptions.
The methods and results need to be statistically valid and make sense from the perspective of common sense.
They need to adequately measure prediction uncertainty, and the measures should be understandable to nonspecialists, such as policy makers.

Currently, most scientific advances focus on some part of these issues. This session seeks to bring scientists together to present their contributions on sensitivity analysis, parameter estimation, and uncertainty assessment, and to develop a vision of how the methods we are all developing can contribute to a truly intelligent way of analyzing prediction uncertainty.


Contact Us General Conference Information Key Note Speakers Special Sessions Presenter Instructions Student Fellow Awards Program Social Event Registration Accomodations Travel & Visa Information Sign up for Email Updates Abstract List CMWR Home	Back to Existing Special Sessions >> Intelligent Analysis of Prediction Uncertainty Conveners: Mary C. Hill, US Geological Survey, and Ming Ye, Florida State University Meaningful evaluations of prediction uncertainty require methods with the following characteristics: They need to be computationally tractable for the long execution times characteristically needed to represent realistically natural hydrologic systems. They need to clearly indicate what aspects of the data and model construction are dominating the calculated measures of uncertainty. They need to indicate how to reduce uncertainty, by collecting more parameter measurements and observation data. They need to account for different types of error, including errors in the data used for model development, errors in estimated parameters, and errors in the conceptual model. They need to be generally applicable, and not limited to rarely satisfied conditions and assumptions. The methods and results need to be statistically valid and make sense from the perspective of common sense. They need to adequately measure prediction uncertainty, and the measures should be understandable to nonspecialists, such as policy makers. Currently, most scientific advances focus on some part of these issues. This session seeks to bring scientists together to present their contributions on sensitivity analysis, parameter estimation, and uncertainty assessment, and to develop a vision of how the methods we are all developing can contribute to a truly intelligent way of analyzing prediction uncertainty.
Privacy & Security Notice \| LBNL Home \| LBNL A-Z Index \| LBNL Phone Book \| LBNL Search \| 1 Cyclotron Rd., MS: 937-0600, Berkeley, CA 94720